People IntelligenceWithout context, AI remains mainly good at formulating answers
A smart model can do a lot with information. But it cannot use context it has never been given.
Björn Slegers ·
The discussion about AI often starts with the model. Which model are you using? How intelligent is it? Can it predict? Can it analyse independently? These are understandable questions. But in HR, perhaps another question is more important. What does the model actually know about the situation you are asking about? Take an absence rate of eight per cent.
Is that high? Perhaps. But for which team? Over what period? After a reorganisation? With how many employees? Has the same manager been there for years? Is this new, or has it been stable for a long time? Without that context, eight per cent is just eight per cent.
People constantly add context
An experienced HR professional sees a figure and almost automatically thinks further. What has changed here? Who is in this team? What happened last month? Is this an exception or a pattern? That does not mean the professional already knows the answer. It does mean that experience helps to understand which questions should be asked first.
AI does not automatically have that organisational context. If Recruitment, contract, performance, Time, absence and reward each contain their own reality, a model sees fragments. It can still write particularly convincing text about them. But a well-formulated answer is not the same as a well-informed answer.
This is why AI makes architecture more important
There is a lot of talk about the quality of large language models. Rightly so. But the ultimate quality of AI support in HR is determined at least as much by what happens before the model. What information is selected? What context is connected? Which data belong to the same employee, role, period or decision?
What information must specifically not be included? And how do we protect identities when external models are used? These are architecture questions. Not just AI questions. A better model does not automatically solve fragmented context.
The best AI does not have to pretend it understands everything
We find it more interesting when AI explicitly helps with preparation. Making a pattern visible. Structuring information from various relevant sources. Signalling a deviation. Preparing questions for a conversation. Placing different scenarios side by side. That can enormously reduce the distance between an initial signal and a human assessment.
But then something happens that is difficult to automate completely. Someone has to determine what it means. Perhaps that is ultimately a better ambition for AI in HR. Not a system that has an answer for everything.
But a system that helps people ask the right question sooner.
How Almuruna One approaches AI and the AI Act